Terms don't end with their definition.

Every entry explains briefly and in depth. Where a term does its work in our articles, the story is linked alongside — that holds for roughly a quarter of the entries today and grows with every article. In an article, the explanation appears right at the word; on a phone, a tap is enough.

100 of 100 terms

01Foundations & ModelsArtificial Intelligence (AI)Software that solves tasks that once required human intelligence.02Foundations & ModelsMachine LearningAI that learns from data instead of being hard-coded.03Foundations & ModelsDeep LearningMachine learning with many-layered neural networks.04Foundations & ModelsNeural NetworkA computational model loosely modeled on the brain.05Foundations & ModelsLarge Language Model (LLM)A large language model that predicts and generates text.06Foundations & ModelsFoundation ModelA broadly pretrained base model that is adapted for many purposes.07Foundations & ModelsFrontier ModelThe most capable, newest top-tier model at any given time.08Foundations & ModelsGenerative AIAI that creates new content (text, images, code, audio).09Foundations & ModelsMultimodal AIProcesses several types of input at once (text, image, audio).10Foundations & ModelsParameterThe learned “knobs” of a model; more usually means more powerful.11Foundations & ModelsModel WeightsThe concrete numerical values of those parameters.12Foundations & ModelsOpen-Weight ModelA model whose weights are freely available.13Foundations & ModelsProprietary ModelA closed model, usable only through the provider or an API.14Foundations & ModelsPretrainingThe first, broad training on huge amounts of data.15Foundations & ModelsFine-TuningPost-training for a specific task or domain.16Foundations & ModelsDistillationA small model learns from a large one (cheaper to run).17Foundations & ModelsQuantizationShrinking a model so it runs faster and cheaper.18Foundations & ModelsReasoning ModelA model that visibly “thinks” before answering.19ArchitectureTransformerThe architecture behind today's language models.20ArchitectureAttentionA mechanism that weights important relations within the text.21ArchitectureDiffusion ModelGenerates images by removing noise step by step.22ArchitectureMixture of Experts (MoE)Only parts of the model are activated per request (more efficient).23ArchitectureEmbeddingConverting text or images into number vectors for the machine.24ArchitectureVector DatabaseA store for embeddings; enables semantic search.25Usage & PromptingPromptThe input or instruction given to an AI model.26Usage & PromptingPrompt EngineeringPhrasing prompts so that they produce good results.27Usage & PromptingSystem Prompta fixed base instruction that frames the model's behavior.28Usage & PromptingZero-Shothaving a model solve a task without examples.29Usage & PromptingFew-Shothaving a model solve a task with a few examples.30Usage & PromptingChain-of-Thoughthaving a model think step by step.31Usage & PromptingTokenthe smallest unit of text models compute with (roughly a word fragment).32Usage & PromptingContext Windowhow much text the model has “in mind” at once.33Usage & PromptingTemperaturea dial for randomness/creativity in responses.34Usage & PromptingHallucinationwhen a model confidently invents something false.35Usage & PromptingGroundingtying answers to verifiable sources/data.36Usage & PromptingRetrieval-Augmented Generation (RAG)the model pulls in matching documents before it answers.37Usage & PromptingContext Engineeringdeliberately providing the model with the right context.38Usage & PromptingGuardrailssafeguards that prevent unwanted outputs.39AgentsAI AgentAI that independently plans steps and uses tools to reach a goal.40AgentsAgentic AIAI systems that act, not just answer.41AgentsTool Use / Function Callingthe model calls external tools/functions.42AgentsModel Context Protocol (MCP)a standard that connects agents with tools and data.43AgentsOrchestrationcontrolling multiple agents/steps as one workflow.44AgentsMulti-Agent Systemmultiple agents working together.45AgentsHuman-in-the-Loopa human reviews/approves at key points.46AgentsAgent Managera person who manages AI agents like employees (job description, limits, audit trail).47AgentsAudit Trailcomplete logging of what an agent has done.48AgentsClosed Loopa system that learns from its results (action → measurement → insight → action).49AgentsWorld Modela continuously maintained, machine-readable understanding of a company/customers.50AgentsIntelligence Layera layer that automatically composes building blocks into solutions.51AgentsCapabilitiesreliable, individual building blocks of ability with no interface of their own.52AgentsWorkflow Automationletting AI handle recurring processes.53Training & AlignmentTraining Datathe examples a model learns from.54Training & AlignmentSynthetic Datatraining data generated by AI.55Training & AlignmentRLHFfine-tuning through human feedback (Reinforcement Learning from Human Feedback).56Training & AlignmentReward Modelauxiliary model that tells good answers from bad ones.57Training & AlignmentConstitutional AIalignment to stated principles rather than to examples alone.58Training & AlignmentAlignmentmaking sure AI does what people want.59Training & AlignmentRed Teamingdeliberately attacking and testing a system to find weaknesses.60Training & AlignmentBenchmarkstandardized test for comparing models (e.g., SWE-bench, MMLU).61Training & AlignmentEval (Evaluation)measuring how well a model solves a task.62Training & AlignmentOverfittinga model memorizes examples instead of generalizing.63Training & AlignmentBiassystematic skew in data or outputs.64Training & AlignmentTests & Harnesseshumans define what “correct” means; AI executes.65Compute & CostComputethe computing power that training and operation consume.66Compute & CostGPUspecialized chip that massively accelerates AI computations.67Compute & CostFLOPmeasure of computing operations (indicator of training effort).68Compute & CostScaling Lawsmore data and compute lead to predictably better models.69Compute & CostInferenceapplying a finished model (the “answering”).70Compute & CostInference Costswhat a single AI answer costs in computing power.71Compute & CostLatencyresponse time until the answer arrives.72Compute & CostTest-Time Computeextra computing time while answering for better “thinking.”73Compute & CostTokenmaxxingstrategy of deploying lots of tokens instead of additional staff.74Compute & CostInference Yieldbusiness value per euro or token spent (rather than sheer volume).75Compute & CostEdge AI / On-DeviceAI runs locally on the device instead of in the cloud.76Compute & CostLocal Modela self-hosted model (data sovereignty, no cloud required).77Business & DeploymentAI Adoptionhow much AI is actually used.78Business & DeploymentCopilotan AI assistant that works alongside a person on the job.79Business & DeploymentAugmentationAI complements humans instead of replacing them.80Business & DeploymentAI-nativea company or product built around AI from the ground up.81Business & DeploymentSell Work, not Softwareselling the outcome or the work instead of a license.82Business & DeploymentAllocation Economyhuman value shifts toward selection, judgment and curation.83Business & DeploymentShadow AIuncontrolled AI use that bypasses IT.84Business & DeploymentTotal Cost of Ownershipoverall costs including operations, not just the license.85Business & DeploymentDisruptionwhen new technology upends existing business models.86Business & DeploymentReskillingtraining employees for new, AI-shaped tasks.87Business & DeploymentProductivity Boostmeasurable efficiency gains from AI.88Business & DeploymentPilot vs. Productiona small-scale test vs. real, continuous operation.89Responsibility, Law & GovernanceResponsible AIresponsible, safe, fair use of AI.90Responsibility, Law & GovernanceEU AI ActEU law that regulates AI by risk.91Responsibility, Law & GovernanceGDPR Complianceusing AI within the bounds of data protection law.92Responsibility, Law & GovernanceModel Carda model’s profile (capabilities, limits, risks).93Responsibility, Law & GovernanceExplainabilitymaking it understandable why the AI decided as it did.94Responsibility, Law & GovernanceDeepfakea convincingly real, AI-generated forgery.95Responsibility, Law & GovernanceWatermarkinginvisible marking of AI-generated content.96Responsibility, Law & GovernanceAI Incidenta harmful event caused by an AI system.97Society, Environment & Big PictureAGI (Artificial General Intelligence)hypothetical AI at broad human level.98Society, Environment & Big PictureEnergy & Water Consumptionpower and cooling needs of large AI data centers.99Society, Environment & Big PictureHonest Signala behavior-based customer signal that does not lie; it accumulates daily.100Society, Environment & Big PictureOpen vs. Closed Sourceopen vs. closed models and ecosystems.

Terms to think further with